The Airflow task execution status is incorrect

[coding-dragon520]

Apache Airflow version

Other Airflow 2 version (please specify below)

If "Other Airflow 2 version" selected, which one?

2.10.4

What happened?

We deployed two schedulers and three workers in a cluster environment, and the task was interrupted in one execution, and the scheduler log showed that the task had failed, but the task was still queued in the metadatabase

What you think should happen instead?

I think it is airflow that produces scenes with inconsistent state in some scenes with high concurrency ...

How to reproduce

This problem is not easy to reproduce, we happened occasionally during production runs, but it was fine when the re-cleark allowed it

Operating System

Ubuntu 20.04.6 LTS

Versions of Apache Airflow Providers

2.10.4

Deployment

Other

Deployment details

using Helm and K8s for installation
Helm：v3.16.2+g13654a5
Kubernates: Client Version: v1.28.14 Kustomize Version: v5.0.4-0.20230601165947-6ce0bf390ce3 Server Version: v1.28.14

Anything else?

No response

Are you willing to submit PR?

• Yes I am willing to submit a PR!

Code of Conduct

• I agree to follow this project's Code of Conduct

[coding-dragon520]

Executor CeleryExecutor(parallelism=500) reported that the task instance <TaskInstance: align_GROUP_525730891524354049.align_152 scheduled__2025-08-29T10:51:40+00:00 [queued]> finished with state failed, but the task instance's state attribute is queued. Learn more: https://airflow.apache.org/docs/apache-airflow/stable/troubleshooting.html#task-state-changed-externally

[coding-dragon520]

But we don't do a task operation on the state, which is reported when airflow is about to execute, For our production problems, if there are workers who are free to help confirm the logic of this piece, see what situations will arise